Card counting
end-to-end.
The Hi-Lo system, true count adjustment, betting spread, index plays, and the brutal honest answer for why none of it works in any crypto-casino blackjack you can sit at right now. The math is the same one Edward Thorp ran in 1962. The application is what changed.
Why counting works at all
Basic strategy assumes a fresh shoe. Every cell in the chart is the right play given a fresh, well-shuffled deck. But as cards get dealt and don't return to the shoe, the composition of the remaining cards drifts away from fresh-deck proportions. That drift is what counting exploits.
Tens and aces favor the player. They make blackjacks (the 3:2 payout), they bust the dealer more often when she's forced to hit 16 down, and they hit the doubles you're making on 10s and 11s. Low cards (2-6) favor the dealer: they let her make 17 from 12-16 without busting.
When a shoe's remaining cards are rich in 10s and aces — i.e. when more low cards have already been dealt than high — the next round is favorable to the player. Counting is just a way to track that drift in real time using mental arithmetic that doesn't look like advantage play to the dealer.
The standard count
Hi-Lo is the canonical entry-level counting system, popularized by the MIT Blackjack Team and used by virtually every advantage player who has come up since. Three card buckets, three values, mental running total.
| Card | Value | Why |
|---|---|---|
| 2, 3, 4, 5, 6 | +1 | Low cards. When the shoe is depleted of low cards, the remaining shoe is rich in 10s and aces — favorable to the player. |
| 7, 8, 9 | 0 | Neutral. These cards have roughly even effects on player and dealer EV. |
| 10, J, Q, K, A | −1 | High cards. When the shoe is depleted of these, the remaining shoe is poor in 10s and aces — favorable to the dealer. |
The whole deck sums to exactly zero (twenty +1 cards, twenty −1 cards, twelve neutrals). That property — "balanced" — is what lets you convert running count to true count by dividing by decks remaining. KO (Knock-Out) drops one rank from the count to make it unbalanced and skip the division step at a small EV cost.
Running count divided by decks remaining
The running count is a single number you carry in your head from the start of the shoe. The true count is what actually controls your bets and your play deviations.
True count = running count / decks remaining Running count = +6 Decks remaining = 3 True count = +6 / 3 = +2 Same running count, different shoe depth: Running count = +6 Decks remaining = 1 True count = +6 / 1 = +6 (much stronger signal)
Why divide? Because a count of +6 with five decks left to deal is much weaker than +6 with one deck left — there are more remaining cards to dilute the imbalance in the deeper shoe. The true count normalizes for shoe depth so a count of +2 in a 6-deck game means the same edge regardless of where in the shoe you're sitting.
Estimating decks remaining is the part most beginners misjudge. The skill is calibrated by tracking the discard tray: a half-full discard tray in a 6-deck game means roughly 3 decks have been played, so 3 decks remain. With practice the eyeball gets within a quarter-deck.
When to bet bigger
A 0.5% counting edge sounds tiny, but it's the weighted-average edge across all true counts. At true count ≤ 0 you're still losing at the basic-strategy rate (~0.4% house edge). The +0.5% edge only emerges when you concentrate your money on the high-count rounds and bet the table minimum at low counts.
Standard betting spread (1-12) True count Bet (units) ≤ +1 1 +2 2 +3 4 +4 8 ≥ +5 12 Higher spreads (1-16, 1-20) bring more EV but more heat from pit bosses. 1-12 is the standard balance for a single counter without team cover. Unit size: usually 0.5%-1.0% of bankroll. $10,000 bankroll → $50-$100 unit → $50 min, $1,200 max when the count is high.
The rule of thumb: at minimum bet you should have enough bankroll to weather a 100-unit downswing without ruin. Anything tighter and a bad shoe runs you out of the game before the long-run edge surfaces. This is the bankroll-management half of advantage play, and it's what separates real counters from people who memorize Hi-Lo and lose anyway.
When the count overrides basic strategy
Most of counting's edge comes from betting bigger when the count is high — basic strategy still tells you the right move on most hands. But at certain true-count thresholds, the basic-strategy play stops being optimal. The 18 highest-EV deviations are the "Illustrious 18" — adding them to a Hi-Lo count adds roughly 0.07% to your edge. The majority of the advantage comes from the first three.
| Hand | Dealer | Basic | Counting | Threshold |
|---|---|---|---|---|
| Insurance | A | No | Yes | TC ≥ +3 |
| Hard 16 | 10 | Hit | Stand | TC ≥ 0 |
| Hard 15 | 10 | Hit | Stand | TC ≥ +4 |
| Pair 10,10 | 5 | Stand | Split | TC ≥ +5 |
| Pair 10,10 | 6 | Stand | Split | TC ≥ +4 |
| Hard 12 | 3 | Hit | Stand | TC ≥ +2 |
| Hard 12 | 2 | Hit | Stand | TC ≥ +3 |
| Hard 11 | A | Hit | Double | TC ≥ +1 |
| Hard 9 | 2 | Hit | Double | TC ≥ +1 |
| Hard 10 | 10 | Hit | Double | TC ≥ +4 |
| Hard 9 | 7 | Hit | Double | TC ≥ +3 |
| Hard 16 | 9 | Stand | Hit | TC ≤ +4 |
| Hard 13 | 2 | Stand | Hit | TC ≤ −1 |
| Hard 12 | 4 | Stand | Hit | TC ≤ 0 |
| Hard 12 | 5 | Stand | Hit | TC ≤ −2 |
| Hard 12 | 6 | Stand | Hit | TC ≤ −1 |
| Hard 13 | 3 | Stand | Hit | TC ≤ −2 |
| Hard 14 | 10 | Hit | Stand | TC ≥ +3 |
The single most valuable index play is insurance at TC ≥ +3: when 10-rich, the chance of the dealer holding a 10 in the hole exceeds the 1/3 break-even point and the side bet flips from a 7.4%-house-edge sucker bet into a +EV play. Skip insurance at any other count.
None of this works on any crypto casino
Counting works because shoe state persists across hands — cards dealt are gone until the dealer reshuffles. That persistence is what every crypto casino specifically eliminates.
Stake Originals Blackjack, BC.Game Originals, and most in-house RNG variants generate a fresh virtual deck for every hand. The shoe state is reset between rounds. There is no persistence to exploit, ever. The provably-fair seed is the only piece of state, and it's rotated per hand by design.
Evolution, Pragmatic Live, Ezugi — every major live-dealer studio runs continuous-shuffle machines or shoe-shuffle machines that re-randomize after every shoe and often after every hand. The deck the dealer is dealing from has no historical memory; the count you've been keeping is reset on the next shuffle, which arrives before any meaningful penetration accumulates.
In a hand-shuffled brick-and-mortar setting, the cut card is placed deep enough that 60-75% of the shoe is dealt before the next shuffle. That deep penetration is what lets the count drift far enough from zero to support a meaningful spread. Online: penetration is effectively 20-30% before the auto-shuffle fires, often much less. The math doesn't have time to work.
The bottom line: no crypto-casino blackjack supports card counting as of 2026. The math is sound; the application is dead online.
Live, hand-shuffled, deep penetration
Counting still pays in person, at brick-and-mortar casinos that hand-shuffle and place the cut card deep. Specific conditions to look for:
- Hand-shuffled shoe — no continuous-shuffle machine, no auto-shoe-shuffler. Pit dealers who shuffle the cards themselves between shoes.
- Deep penetration — cut card placed roughly 75% through the shoe (so 4.5 decks out of 6 are dealt before the next shuffle).
- 3:2 blackjack payout — never count at a 6:5 table; the payout drop more than wipes out the counter's edge.
- Bankroll discipline — at least 200 max-bet units in reserve. Variance at a spread of 1-12 is significant; thinly-bankrolled counters get blown out before edge surfaces.
- Cover — counting is legal everywhere, but private casinos can refuse service, ban you indefinitely, or subject you to closer scrutiny once they suspect. Camouflage techniques (varying spread, occasional "mistakes", mixing in non-counted sessions) are part of the actual job.
Treat counting as cool theory or a long-game project for live travel. As an internet-era edge play, it stopped paying when continuous-shuffle machines hit the floor in the late 1990s and finished dying when crypto casinos standardized RNG-per-hand reshuffling in 2018.
Beyond Hi-Lo
Hi-Lo is the standard, but a handful of more sophisticated systems exist for serious counters who want to extract the last fraction of a percent from the game. The trade-off is always the same: more accuracy at higher mental load. For recreational counters, Hi-Lo captures ~95% of the available edge with a fraction of the cognitive cost.
What to read next
Variant-aware chart with per-cell EV tooltips. Master this before you touch counting — counting on top of bad basic strategy is a net loss.
How rule changes shift the math. Counter-relevant: 6:5 payouts and CSMs are deal-breakers regardless of how good your count is.
How the game works, what the dealer is doing, and where the math actually lands.